Handle Race Conditions / Concurrency in Code First Entity Framework Applications


A common scenario in developing applications is the handling of concurrent database updates.

Given there is an application to manage customers, that is used by multiple users simultaneous.

User U1 reads the data of customer C1. While user U1 is looking at this data, user U2 reads the data of customer C1 too. Meanwhile, user U1’s phone is ringing. She picks it up. User U2 changes some of the customer’s data and saves it to the database. User U1 finishes the phone call, make some changes to the same customer, and saves it too.

Without handling the race condition or concurrency, al changes made by user U2 will be lost in this example. Handling it, the application should not save the changes made by user U1 and inform her that the customer’s data was changed in the meantime by another user.

In this post I will show an approach and sample code on how an application using the Entity Framework and the Code First approach can handle this.

Code Preparation

There is only one step required to make Entity Frame do all the work for you: Define a Version property in the model class (the name of the property does not matter at all), and decorate it with a Timestamp attribute.

/// <summary>
/// This column will be used by EF for race condition validation.
/// </summary>
public byte[] Version { get; set; }

Believe it or not – this is all you have to do (in case you are using the Code First approach).

Entity Framework will set the Version property to a value when the record is inserted into the database. Every time EF updates the record, it verifies that the value of the column has not changed. Doing an update, the version is increased.

Handling a Race Condition

EF throws a DbUpdateConcurrencyException in case a record was changed since the data was read.

So you have to catch this exception whenever your application updates or deletes database records and inform the user that the data was changed. In case there is no user to be informed, e.g. if this happens in a background process, you have to implement an appropriate error handling mechanism.

Using the Sample

To see EF Concurrency Handling in action, you can use the sample application.

There are two “clients” implemented, accessing both the same database record. Take these steps to provoke a concurrency exception:

  • Read the record for client 1 (press the Read Record button of the Client 1 group)
  • Read the record for client 2 (press the Read Record button of the Client 2 group)
  • Update the record for client 2 (press the Update Record button of the Client 2 group)
  • Try to update the record for client 1 (press the Update Record button of the Client 1 group)

The result should look like this:

Sample when Race Condition occurred

Some Sample Details

The sample code was created using Visual Studio 2013 Ultimate and Entity Framework 6.1. In case you use an older version of Visual Studio, you might have to create an empty solution and add the files to it.

The application expects a SQL Server instance (not SQL Server Express) installed on the local machine having the default server name (MSSQLSERVER). In case you do not have a SQL Server with this name running on your machine, you have to change the connection string in the app.config file before you can run the sample.

On the SQL Server, the application creates a database named EntityFrameworkRaceCondition. Please make sure to have the appropriate rights to create a database when you run the sample.


Microsoft Data Developer Center Entity Framework

Entity Framework / Get Started / Code First to a New Database

Entity Framework Tutorial Update Entity Using DbContext

Wikipedia Race Condition

Sample Application

Disk Space Side Effects of SQL Server Data Tools

SQL Server Data Tools (SSDT) installs LocalDB, a new version of SQL Express. For details on LocalDB, please refer to Introducing LocalDB, an improved SQL Express.

As Kevin Cunnane explains in the SQL Server forum thread Importing database always defaults to localdb, creating a SQL Server database project by SSDT always creates a database on LocalDB, (localdb)\Projects since September 2012 update of SSDT. And of course, one cannot change this behavior – well, why should we, just because we do have a SQL Server installed on our dev box?

Kevin points out that the database is empty and only populated on debug/deploy. So another 4 MB are wasted on my profile drive. Why on my profile drive? Because the database files are located on %user%/AppData/Local/Microsoft/VisualStudio/SSDT. Of course, you can delete the database, also from within SQL Server Management Studio, as soon as you close the database project. But when you re-open the project, the database is re-created.

The system databases, error logs and event files are located at %user%/AppData/Local/Microsoft/Microsoft SQL Server Local DB/Instances/Projects (or V11.0, if you do not have September 2012 update or later installed). Here you can remove the event log and system health files from time to time, if LocalDB won’t do so. If you like to test it (I didn’t): the directory is set in the registry under HKEY_CURRENT_USER\Software\Microsoft\Microsoft SQL Server\UserInstances. In case there is more than one entry below this key, have fun 😉 This seems to happen when you have SSDT installed before the September 2012 update, and then updated it.

Trying to change some settings of LocalDB using SQL Server Management Studio, like database default locations or error log recycling, led to an ‘Access denied’ exception on my machine.

But that’s not all. There is also a .dbmdl file created, having about 8 MB on my really small sample database project. This file can be deleted too, but will be re-created. Some guys on the web say that .dbmdl files are only used as a cache per user. This is why they should not be put under source control.

So in case you run out of disk space, check your profile for LocalDB database, log or health files and your projects for .dbmdl files.

Where Do SSDT SQL Scripts Connect To?

In some software projects I used empty Visual Studio solutions to create some kind of database projects. I added SQL script files to the solutions to create all database objects, like tables, functions, stored procedures, views, constraints, and so forth. A batch file was used to run all these scripts against the database.

One common part of these script files was a snippet of sample code to test the database object, e.g. a stored procedure or function, directly from inside Visual Studio. This snippet both documented the use and gave a quick way to test changes. Of course, this snippet was intended to be used only when connected to a local SQL instance to avoid damage in any other environment!

Playing around with SQL Server Data Tools (SSDT) in VS 2012, I was looking for the ability to connect a function script to my existing local database. Under the menu item SQL / Transact SQL Edit / Connection there is the ability to connect to a database. To my surprise, I was not asked to which SQL instance or database I would like to connect to. At least, clicking the connect menu item seemed to do nothing at all.

But running the script worked, and the result window told me that the function was created successfully. I was wondering where. Searching for a database selection option in the toolbars, as I know it from the SQL script files in VS 2010, failed. There is no such possibility. Was the function created in the master database? But which SQL Server instance was used?

Well, Visual Studio showed me the answer to these questions all the time, but I did not noticed it.

When you edit a function or a stored procedure script file of a SQL Server Database project – not a simple SQL Query file! – you will find these information in the properties window on the right (or wherever you have placed it). The properties window shows the connection name, logon name and so forth, and whether the connection is open or closed. You cannot change any entry. These properties are not available for table script files of the project, which can’t be connected to any database at all from within Visual Studio.

But the properties window does not tell you the name of the database. Almost, it is the project’s name. But it doesn’t have to. The name can be found in the .sqlproj.user file, located in the project directory. Of course, you can edit this file to change the name.

Create a SQL Azure Database using SSDT

To get into Microsoft Azure application development, I was playing around with SQL Server Data Tools (SSDT), trying to create a database on SQL Azure.

I started creating a simple database, having two tables with a few columns. As done before with other tools, I was using the extended properties to document the database ‘inline’. The target platform was my local SQL 2012 instance. Deployment went fine, updating after some changes as well.

So the SQL Azure database server was created quickly, and I tried to deploy that database from my local machine to SQL Azure. It failed – of course (I wouldn’t have written a post just to tell you I succeeded 😉 ).

The reason surprised me, but a solution was found very quickly.

The reason why publishing failed is, that, beside others, SQL Azure does not support extended properties.

The solution is posted by Bill Gibson on the SQL Server Data Tools Team Blog in his post Migrating a Database to SQL Azure using SSDT.

I created a Schema Compare file as suggested by Bill, and put a second SQL Server project into my solution, following his instructions – and it worked fine 🙂

Now I have two database projects in my solution. The first one with full featured extended properties to document the database by itself. This one is used to build and update the local SQL 2012 version of the database. And the second one, based on Schema Compare, to be able to convert the SQL 2012 version into a SQL Azure compatible one and build and update the instance on SQL Azure. Of course, one have to take care not to update the scripts of the SQL Azure version, but keep in mind that SQL 2012 is the master.

This works smooth and easy in my simple test environment, because I do not use such things as user defined types (CLR) or XML indexes. Both are also not supported by SQL Azure. And migrating an existing database which uses these feature is not just letting Schema Compare doing the work. In such a case, one has to do some re-work, or probably re-think the decision to move to SQL Azure.

Btw, I think it’s worth to mention that when excluding the Default objects from schema comparison, this does not mean that column default values created by the CREATE or ALTER TABLE statement will be ignored. It seems that only those defaults created by CREATE DEFAULT are ignored. To me this is OK, since the SQL Server books say that CREATE DEFAULT will be removed in future versions of SQL Server anyway.

machine.config Overrides TransactionScope / TransactionOptions.Timeout

You are wondering why your DB-transaction, startet by

using(TransactionScope scope = new TransactionScope(timeout))

runs into a timeout after 10 minutes?

Because the

<machineSettings maxTimeout="00:10:00"/>

in the system.transactions section of the machine.config has a default value of 10 minutes. And whatever is set as timeout of the TransactionScope by code – machine.config overrides it all.

So you have to change the machine.config, which does not contain a system.transaction section by default – which doesn’t mean the timeout default value does not exist.

And as long as you hadn’t set a timeout in the code, you don’t have to change any line. The new value of the machine.config is used when the process starts. No machine reboot is required.

Btw: You will find the machine config in %install path%/Config. Which is usually c:\windows\Microsoft .NET\Framework\V…\Config. On x64 machines, there is a Framework64 directory as well. My suggestion is to change all machine.config files to make sure every environment behaves in the same way.

The source of my knowledge: http://technet.microsoft.com/en-us/library/ms149852(VS.90).aspx

SQL Server Isolation Level Snapshot & DTC Promotion

SQL Server transaction isolation level Snapshot cannot be promoted by DTC.

In case you open a transaction scope in .NET code, using isolation level snapshot, then open a connection, leave that connection open, and try to open another connection, you will retrieve an exception saying that a snapshot transaction cannot be promoted.

Well, in this case, you didn’t wanted to promote that transaction. Using the snapshot isolation level, just make sure only one connection is open at a time, and no exception will occur 🙂

XML Data Row Truncation At 2,033 Chars When Using SqlDataReader

When you read Extensible Markup Language (XML) data from Microsoft SQL Server by using the SqlDataReader object, the XML in the first column of the first row is truncated at 2,033 characters. You expect all of the contents of the XML data to be contained in a single row and column.

This behavior occurs because, for XML results greater than 2,033 characters in length, SQL Server returns the XML in multiple rows of 2,033 characters each.

To resolve this problem, use the ExecuteXmlReader method to read FOR XML queries. For additional information about how to use ExecuteXmlReader with SQL Server FOR XML queries, click the article numbers below to view the articles in the Microsoft Knowledge Base:
316016  (http://support.microsoft.com/kb/316016/ ) How to use the ExecuteXmlReader method of the SqlCommand class in Visual Basic .NET

316701  (http://support.microsoft.com/kb/316701/ ) How to use the ExecuteXmlReader method of the SqlCommand class in Visual C# .NET

307224  (http://support.microsoft.com/kb/307224/ ) How to use XML in connected and disconnected ADO.NET application

or: Use SqlDataReader an iterate through all the lines.